review

Review code diffs, patches, and pull requests for defects and merge readiness.

5|Updated Jan 31, 2026
One-click install
npx skills add https://github.com/nuggocto/dotfiles --skill review-nuggocto
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: review
Source: https://github.com/nuggocto/dotfiles/tree/main/opencode/skills/review
Command: npx skills add https://github.com/nuggocto/dotfiles --skill review-nuggocto

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code reviews often miss real defects or drown in nitpicks and unverified suspicions. This Skill enforces an adversarial, evidence-first review process that confirms every finding against the actual code before reporting it, producing an accurate merge decision. ## Core Features & Use Cases - Adversarial verification: Starts from the hypothesis that the change contains a serious defect, then tries to falsify each suspected issue before reporting it. - Contract-first analysis: Establishes requirements, backward compatibility, and scope before judging the implementation, so intended behavior changes are not mislabeled as regressions. - Structured verdict reporting: Leads with confirmed findings ranked by severity, each with file/line location, violated contract, failure path, impact, evidence, and a minimal fix direction. - Use Case: Ask it to review a pull request before merging; it inspects the full diff, traces failure paths, runs safe read-only checks, and returns either confirmed findings or an explicit clean pass. ## Quick Start Ask the agent to review the current branch diff or a specific pull request and report confirmed findings with severity and merge readiness.

Frequently Asked Questions about review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a pull request for defects before merging?

Point the review at the full diff including staged, unstaged, and untracked files. It establishes the intended contract, traces failure paths through callers and guards, and reports only findings confirmed against the actual code.

What makes a code review evidence-based instead of opinion-based?

Each finding must cite the violated contract or invariant, the concrete trigger conditions, and the traced execution path. Suspicious patterns and scanner output are not reported until their impact is verified against the repository.

Does the review skill modify my code while reviewing?

No. Review is read-only by default; it never edits source, changes configuration, or rewrites tests. It may run safe local read-only checks, and remediation happens only when explicitly requested.

What happens when a code review finds no defects?

It returns an explicit clean verdict stating no confirmed findings in the reviewed scope, plus any meaningful verification gaps or residual risks. It does not invent issues to avoid a clean pass.

How are severity and confidence handled in review findings?

Severity describes user or system impact, while confidence describes the strength of evidence, and neither is inflated to compensate for the other. Unverified items are labeled as questions or risks, not findings.